User Researcher, AI Evaluations

We’re seeking an experienced UX Researcher to define and scale how we evaluate Notion’s AI-powered experiences—focusing on what “good” looks like not only for model output quality, but for the end-to-end product experience where people discover, set goals, delegate work, review results, and build trust over time with AI.

This role sits at the intersection of research craft and evaluation operations: you’ll run studies that uncover user mental models, expectations, and failure/recovery behaviors, then translate those insights into reusable rubrics, workflows, and measurement approaches that product, design, engineering, and data science can apply consistently.

This role can be based in either San Francisco or New York City. We work from our offices on Mondays, Tuesdays and Thursdays (our Anchor Days) because we do our best thinking and building together in person. We’re looking for someone who’s excited to work alongside the team during those days.

What You'll Achieve:

• Define what “good” looks like (frameworks & rubrics): Establish clear, reusable evaluation criteria that reflect real user expectations—helpfulness, trust, tone, control, and transparency. You’ll translate qualitative insight into scoring guidance that can be applied consistently across teams and over time.
• Run recurring evals (longitudinal & feature-specific): Run recurring longitudinal and feature-specific surveys and studies to measure experience quality over time against defined rubrics. Lead qualitative studies, side-by-side comparisons, and human-in-the-loop evaluation efforts to deepen understanding of where experiences break down and how they can improve. You’ll help teams spot regressions, benchmark improvements, and understand when expectations shift.
• Anchor evaluation in real workflows (context > isolated feedback): Ensure evals reflect jobs-to-be-done, user intent, and the full interaction journey (goal setting, delegation, review, iteration), not just decontextualized thumbs up/down. You’ll help teams understand who is evaluating, what they’re trying to do, and why outputs succeed or fail.
• Identify failure modes & recovery behavior (guardrails): Uncover breakdowns, regressions, and edge cases across the system—from model behavior to UI and integrations—and study how people notice issues, correct them, and continue their work. You’ll turn these insights into actionable guidance for guardrails, fixes, and prioritization.
• Operationalize evaluation with partners (process & tooling): Collaborate closely with Product, Design, Engineering, and Data Science to align on target use cases and build scalable evaluation loops (human-in-the-loop review, longitudinal studies, and calibration of automated/LLM-judge approaches against human judgment).